We devise a technique to remove texturing artefacts that are typical of 3D models acquired by photogrammetric techniques. The system constructs a local parametrization P around the texture defect, fills its domain with the texture data, and performs a context-aware inpainting operation using a neural network.

We propose a novel line drawing vectorization framework based on coarse-to-fine curve network optimization. Our vectorization tool is able to produce high-quality curves that are faithful to the original inputs and close to the connectivity of human drawings.

This paper describes a GPU-CPU hybrid parallel algorithm for computing the extremum graph of scalar fields in all dimensions. An open source software library, TACHYON, that implements the algorithm exhibits superior performance and good scaling behaviour. Extremum graph for the silicium grid (left) and gradient paths between maxima and 2-saddles that constitute arcs of the graph (right).

Labels and clusters can be used to leverage the computations to identify and group point sets with similar structure.

• The refined in each cluster is used to overcome the distribution irregularities of points.
• This statistical-based non-rigid point cloud registration approach can address the challenges of connections and separations caused by object deformation and large inter-frame motions

We propose a unified feature representation for high-resolution clothed human reconstruction, which integrates the human shape feature representation and the details feature representation to finish high-quality reconstruction for clothed human with arbitrary poses and clothing details.

This work introduces a new loss function grounded in spectral geometry and applicable to different neural-network-based generative models of 3D head and body meshes. A model trained with our local eigenprojection loss (1), can be used to generate and edit human shapes by directly manipulating the latent variables (2).

We propose a colorization approach for cartoon sketches, which takes both sketches and color hints as inputs to produce impressive images.

Unified points, beams and paths (UPBP) is a light transport algorithm that is capable of simulating light effects in arbitrary input scenes that contain participating media. UPBP uses bidirectional path tracing and photon density estimation to form full light paths by combining subpaths from light sources with subpaths the camera. Multiple importance sampling is used to compute the weight of these light paths, which is a computationally expensive task. We derive a new algorithm to more efficiently compute this weight, which improves over previous work by eliminating the need to iterate over all path vertices.

We propose an approach to acquire high-fidelity 3D models of off-the-shelf clothing using a mannequin with simple setup. Extraction of both tight and loose garments from clothed mannequin scans is enabled with a single pipeline.

We introduce double trees as a visualization approach to abstract and compare event sequences. The approach aggregates the sequences before and after an interactively selected event of interest. We extend the approach to contrast event sequences discerned into colour-coded categories by user-defined criteria.

This paper conducts a systematic survey on the recent progress of personalized interior design (PID) from furniture selection and floor plan preparation. The former selects furniture in a consistent style according to user preference, while the latter generates a reasonable floor plan design according to the house structure.

A complex dynamic phenomenon may consist of heterogeneous components with diverse patterns of changes over time. Our approach to producing a unified periodization for multiple heterogeneous components is designed for situations where the user wants to understand the overall behaviour of the phenomenon as a whole by obtaining an easily understandable representation. The approach involves a combination of computational and interactive visual techniques that support division of time into meaningful and manageable periods enclosing different relatively stable states or development trends, which may include creation and subsequent integration of multiple different divisions.

We introduce MesoGAN, a model for generative 3D neural textures. This new graphics primitive represents mesoscale appearance by combining the strengths of generative adversarial networks (StyleGAN) and volumetric neural field rendering. The primitive can be applied to surfaces as a neural reflectance shell and rendered in modern path tracers.

We present a new continuum simulation framework in human-solution space for realistic crowd motion generation. To leverage the advantages of model-based and data-driven approaches, a multi-granularity physics-based model is designed to be data-friendly. To achieve better realism, an acceleration-aware data-driven optimization scheme is proposed to mimic real-world motion dynamics.

This paper presents a novel end-to-end network architecture for portrait-background composition. The method adjusts the appearance of portraits to make them compatible with backgrounds, while the generation of the composited images satisfies the prior of a style-based generator. The proposed method outperforms other state-of-the-art methods on the synthesized dataset and the real composited images and shows robust performance in video applications.

We propose a neural network-based character locomotion stitching technique with recurrent motion refiner (RMR). We created paired data through novel data generation that employs K-nearest neighbour search and trained our network to connect discontinuous locomotions into a single natural locomotion.

This research introduces a human-in-the-loop framework, EvIcon, designed to improve the usability of interface icons. The framework uses a refined version of a large-scale pre-trained joint image-text embedding (IconCLIP) and a new dataset, IconCEPT10K, to provide instant feedback on icon usability and visual distinguishability. The proposed framework enhances the icon revision process and results in interface icons with better semantic distance and familiarity.

We applied a theoretical model to develop an abstract general approach to analysing episodes (prolonged events) characterised by multiple time-variant attributes. It involves incrementally increasing the level of data abstraction by merging multiple elements into patterns. In an example implementation, we used topic modelling to obtain multi-attribute variation patterns.

This work presents a novel implementation of the screened Poisson surface reconstruction algorithm that allows the reconstruction to be distributed across multiple clients using a client-server system, with only a small number of synchronization barriers. By decomposing the solution of the linear system into low-frequency (global/server) and high-frequency (local/client) components, leveraging padding, and enforcing a connected isosurface, we obtain a solution that exhibits no artifacts at client boundaries and is indistinguishable from the single-client solution.

We propose a semi-procedural differentiable material prior that represents materials as a set of grayscale noises and patterns that are processed by a sequence of lightweight learnable convolutional filter operations. Combined with a differentiable rendering, we enable single-image tileable material capture comparable with state of the art.

We propose to use nonlinear shape functions represented as neural networks in numerical coarsening to achieve generalization capability as well as good accuracy. To overcome the challenge of generalization to different simulation scenarios, especially nonlinear materials under large deformations, our key idea is to replace the linear mapping between coarse and fine meshes adopted in previous works with a nonlinear one represented by neural networks.

This paper presents two-step training and spatial latent manipulation based on a pre-trained Convolutional Neural Network and self-adaptive Multilayer Perceptron. The proposed model can achieve high-quality and color-adjustable results using reference images and text tags.

Reference-based screentone transfer results by our method. Given a line drawing (i.e. input), and a reference manga image, our model can transfer the screentone pattern from the reference to the line drawing. We show each group of the results with three different reference manga images per row.

With the occlusion-aware feature representation, the proposed method can improve the quality of re-constructed geometry and pose robustness for single-view clothed human re-construction.

We present a simple and effective method for the interactive segmentation of feature regions in a triangular mesh. The experimental results show several segmentation results for various 3D models, revealing the effectiveness of the proposed method.

We proposed an intelligent companion learning system named IARE, which seamlessly integrates AR and AI based on the ARCS model, trying to stimulate the role of parents and provide an intelligent and companion-based learning experience for children learning by themselves.

In this review, we survey the literature focusing on molecular visualization in immersive environments. We report on various enabling technologies, such as head-mounted displays and augmented and mixed reality. Furthermore, we identify key domains, use cases, and visualization and interaction techniques employed by the current research.

We present Evonne as a comprehensive web tool for explaining reasoning through interactive visualizations of proofs and ontologies. Evonne uses specialized views for exploring explanations of logical consequences and the knowledge from which these are derived, additionally supporting repair of ontologies in case of erroneous entailments.

In our survey of visual parameter space exploration with spatial/temporal data, we identify five themes forming the components of a common workflow.

S-EPB computes an edge-path bundling of a straight-line drawing (a) by bundling along paths in a t-spanner (b), a sparse sub-graph, while still avoiding edge ambiguities in the bundled drawing (c).

In this paper, we aim at providing researchers and practitioners alike with a roadmap detailing the current research trends in the field of network visualization. We design our contribution as a meta-survey where we discuss, summarize, and categorize recent surveys and task taxonomies published in the context of network visualization.

This paper introduces a new multilevel robustness framework for studying time-varying vector fields, which can differentiate the behaviours of critical points in terms of their multiscale stability. This framework supports feature tracking, selection and comparison, and improves the visual interpretability of vector fields from scientific simulations.

iFUNDit is an interactive visual analytics system for fund investment style analysis. Itdecomposes a fund's critical features into performance attributes and investment stylefactors, and visualizes them in a set of coupled views: the Fund View and Manager View, todelineate the distribution of funds' and managers' critical attributes on the market; the Cluster view, to show the similarity of investment styles between different funds; and theDetail View, to analyze the evolution of fund investment style. The system provides a holisticoverview of fund data and facilitates streamlined analysis of investment style at both thefund and the manager level. The effectiveness and usability of the system are demonstrated through domain expert interviews and case studies by using a real mutual fund dataset.

We combined and adapted three existing design spaces for visual data stories to classify 130 spatio-temporal stories collected between 2018 and 2022. An analyzis of the collected data revealed various patterns, for example how large-scale struggles shape the development of storytelling techniques.

This paper presents new techniques for seamlessly transitioning between parallel coordinate plots, star plots, and scatter plots. The design led to a variant of the star plot with curved connections between axes and a geometrically motivated embedding of scatter points from a scatter plot into star and parallel coordinate plots.

We present a structure to efficiently query the Signed Distance Field (SDF) of triangle meshes. The method stores the triangles influencing the field behaviour in each region of an adaptative grid to accelerate the nearest triangle search task later.

Our findings demonstrate that the original ARAP approach can be construed as minimizing a discretization of an elastic energy that is based on non-conforming elements defined over dual orthogonal cells of the mesh. By utilizing intrinsic Voronoi cells instead of an orthogonal dual of the extrinsic mesh, we ensure that the energy remains non-negative within each cell. We depict the intrinsic Delaunay edges as polylines over the mesh, represented in barycentric coordinates relative to the mesh vertices. This modification of the original ARAP energy, which we refer to as iARAP, resolves issues arising from non-Delaunay edges in the original method. In contrast to the spokes-and-rims version of the ARAP approach, it is less sensitive to the triangulation of the surface.